Irreversible disability in multiple sclerosis (MS) and neuromyelitis optica spectrum disorder (NMOSD) is largely attributed to neuronal and axonal degeneration, which, along with inflammation, is one of the major pathological hallmarks of these diseases. Optical coherence tomography (OCT) is a non-invasive imaging tool that has been used in MS, NMOSD, and other diseases to quantify damage to the retina, including the ganglion cells and their axons. The fact that these are the only unmyelinated axons within the central nervous system (CNS) renders the afferent visual pathway an ideal model for studying axonal and neuronal degeneration in neurodegenerative diseases. Structural magnetic resonance imaging (MRI) can be used to obtain anatomical information about the CNS and to quantify evolving pathology in MS and NMOSD, both globally and in specific regions of the visual pathway including the optic nerve, optic radiations and visual cortex. Therefore, correlations between brain or optic nerve abnormalities on MRI, and retinal pathology using OCT, may shed light on how damage to one part of the CNS can affect others. In addition, these imaging techniques can help identify important differences between MS and NMOSD such as disease-specific damage to the visual pathway, trans-synaptic degeneration, or pathological changes independent of the underlying disease process. This review focuses on the current knowledge of the role of the visual pathway using OCT and MRI in patients with MS and NMOSD. Emphasis is placed on studies that employ both MRI and OCT to investigate damage to the visual system in these diseases.

Multiple sclerosis lesion activity segmentation is the task of detecting new and enlarging lesions that appeared between a baseline and a follow-up brain MRI scan. While deep learning methods for single-scan lesion segmentation are common, deep learning approaches for lesion activity have only been proposed recently. Here, a two-path architecture processes two 3D MRI volumes from two time points. In this work, we investigate whether extending this problem to full 4D deep learning using a history of MRI volumes and thus an extended baseline can improve performance. For this purpose, we design a recurrent multi-encoder-decoder architecture for processing 4D data. We find that adding more temporal information is beneficial and our proposed architecture outperforms previous approaches with a lesion-wise true positive rate of 0.84 at a lesion-wise false positive rate of 0.19.